Title: Multiscale atmospheric simulation using the spectral element method Aim Fournier NCAR also with F' B
1Multiscale atmospheric simulation using the
spectral element methodAimé Fournier
(NCAR)also withF. Baer, Houjun Wang (UMCP)A.
Pouquet, D. Rosenberg, J. Tribbia (NCAR)Mark
Taylor (Sandia)
2Why Do This
- Variable resolution in a model helps define
climate - Nonlinear effects introduced by regional scales
must be incorporated into a climate - Smaller scale effects often grow on shorter time
scales - Identification and prediction of regional climate
should help in understanding the evolution of the
global climate. - Integrations must be sped up to perform all
computations needed for solving the climate
modeling problem.
3SEACM Spectral Element Atmospheric Climate Model
What is the model?
- A global model with an unstructured grid and some
useful features - Uses geometric properties of finite element
methods - Incorporates local mesh refinement and regional
detail - Takes advantage of parallel processing
- Maintains the accuracy of spectral models
- Is computationally efficient
- Has no pole problem.
4Setting The Model Domain
- Tile the spherical surface with arbitrary number
and size of rectangular elements - Inscribe a polyhedron with rectangular faces
inside sphere, - Map surface of polyhedron to surface of sphere
with a gnomonic projection, - Use the cube (most elementary polyhedron),
- Subdivide each of the six faces of the cube as
desired. - Can use Local Mesh Refinement (LMR) as desired.
5Subdivision 1
Cube
Uniform Resolution Rectangles
6Quick Summary of the Integration Process
- Represent the prediction equations in integral
form - Use Gauss-Lobatto quadrature for integration
- Use Legendre cardinal functions for the basis
functions - Use test functions based on the Legendre cardinal
functions. - These choices result in an extremely simple
finite element method with a diagonal mass
matrix. -
7Examples of Picture Framing
Basic grid
Expanded grid (3x3)
LMR on the globe
8Topography w/o LMR
T42
T85
Note Each element has an 8x8 grid
Note Each element has an 8x8 grid
9Topography with LMR over Andes
T42
3x3
9x9
Note Each element has an 8x8 grid
10Some test experiments
- Experiments without physics
- Run both CAM/EUL and CAM/SEACM without physics
(only DC) with H-S forcing, topography and real
ICs - Run the cases for 10 days with uniform
resolution, T42 for both models, with the ICs
from NCAR documentation - These experiments should give an indication of
the quality of SEAM predictions.
11Sea Level Pressure (hpc)
Dynamical cores W/o topo.
Seam
T 0 days
EUL
T 5 days
T 10 days
12Sea Level Pressure (hpc)
Dynamical cores with topo.
Seam
T 0 days
EUL
T 5 days
T 10 days
13More test experiments
- Experiments with physics
- Run both CAM/EUL and CAM/SEACM with physics,
topography and real ICs - Run the cases for 10 days with uniform
resolution, T42 for both models, with the ICs
from NCAR documentation - These experiments should give an indication of
the quality of SEACM predictions.
14Sea Level Pressure (hpc)
CAM/Swamp w/o topo.
Seam
T 0 days
EUL
T 5 days
T 10 days
15Some computation and timing results
- Results are with DC SEAM and H-S forcing
- Computations with various truncations and of
processors - Computations with various computers and of
processors.
16Dynamical Core/SEAM
- Held-Suarez forcing
- SEAM with uniform grid
- Scaling results for various resolutions almost
insensitive to processor number change.
HP Exemplar SPP2000
320km/L20 160km/L20 80km/L20
320km/L20 (dotted) 160km/L20 (solid) 80km/L20
(dashed)
17Dynamical Core/SEAM
- Parallel scaling on various computers
- Log-log plot, flops vs processors.
Gflops
- Triangles denote SEAM
- Horizontal resolution-T181, (g)seaborg- T533
- Aries DyCore
- Other symbols for other models
of processors
18Breakdown of the Polar Stratospheric Vortex
- 16-day simulation with SEAM
- Highest resolution case
- Horizontal Resolution 36 km (T363)
- Vertical Resolution 200 levels
- Domain global
- Supercomputer IBM SP RS/6000
- Grid Points 88780800
- Results from simulations with SEAM and other
models for coarser resolutions are available.
19Polar vortex calculation comparisons
Cost per day per level
Ger. Icos. model
Eulerian
Finite-vol
Semi-Lag.
SEAM
20- Polar vortex evolution
- potential vorticity
- medium resolution(70 km)
- view from space
- 16 days
21- Polar vortex evolution
- potential vorticity
- medium resolution (70 km)
- view from space, exaggerated height
- 16 days
22- Polar vortex evolution
- potential vorticity
- high resolution (36 km)
- view from space,exaggerated height
- 16 days
23- Interactions
- NCAR
- CCM4 staff
- SCD staff
- Scientists associated with the project
- Computers
- NERSC
- Computers and Support staff
- ORNL
- Computers and Support staff
- UMCP
- Scientists and students associated with the
project - Stretched-grid development group.
- Staff
- PIs Baer, Tribbia, Fournier
- Postdoc Wang
- Co-Investigator Taylor
- Faculty Affiliate Fox-Rabinovitz
- Collaborators Thomas, Loft
24The End